docs(v0): nullsec-s1 security-LLM integration plan (refs #67) #79

Merged

pdurlej merged 1 commit from gemini/issue-67 into main

2026-06-08 21:46:31 +02:00

claude commented

2026-06-08 21:41:47 +02:00

Collaborator

Authored by gemini (Gemini 3.5 Flash via Antigravity), Swarmheart worker under claude's arbitration. claude reviewed the full doc + corrected one factual error (see below).

What

Planning doc docs/operations/nullsec-s1-integration-plan.md for evaluating nullsec-s1 (open-source security-specialized LLM) integration. PLAN, not execution — parked under M2 per D21, sibling of pyfallow-integration-plan.md. Refs the #67 triage.

Content (claude-reviewed)

Mirrors the pyfallow-plan structure: status banner, why, options A/B/C, recommended path, decision triggers, pre-conditions, wrongness signals, M3 open questions.
Both hard caveats prominent + correct:
- Data egress — default hosted backend (s1.trynullsec.com) ships code off-box → must default-block egress, self-host or explicit opt-in.
- Sensor-not-gate (D20) — nullsec's --fail-on is a gate; Patchwarden must consume findings + let policy decide, never let the external tool block merge.
Option A (CI-subprocess) recommended; B (Python dep) rejected (stdlib-only); C (self-host) borderline.
Honest: Apache-2.0 clean, complementary to pyfallow (structural) + Ollama (generic).

Arbiter note (claude)

Fixed one factual error before shipping: Flash wrote the repo org as github.com/nullsec-ai/...; the authoritative link (issue #67 + the trynullsec.com backend) is github.com/trynullsec/nullsec-s1. Corrected, folded into gemini's commit.
Scope: docs-only, single new file (+179). 220/220 tests still green (no code touched).
Authored by gemini (identity isolation held).

Refs #67. Sibling: pyfallow-integration-plan.md. D20/D21.

> **Authored by gemini** (Gemini 3.5 Flash via Antigravity), Swarmheart worker under claude's arbitration. claude reviewed the full doc + corrected one factual error (see below). ## What Planning doc `docs/operations/nullsec-s1-integration-plan.md` for evaluating `nullsec-s1` (open-source security-specialized LLM) integration. **PLAN, not execution** — parked under M2 per D21, sibling of `pyfallow-integration-plan.md`. Refs the #67 triage. ## Content (claude-reviewed) - Mirrors the pyfallow-plan structure: status banner, why, options A/B/C, recommended path, decision triggers, pre-conditions, wrongness signals, M3 open questions. - **Both hard caveats prominent + correct:** - **Data egress** — default hosted backend (`s1.trynullsec.com`) ships code off-box → must default-block egress, self-host or explicit opt-in. - **Sensor-not-gate (D20)** — nullsec's `--fail-on` is a gate; Patchwarden must consume findings + let *policy* decide, never let the external tool block merge. - Option A (CI-subprocess) recommended; B (Python dep) rejected (stdlib-only); C (self-host) borderline. - Honest: Apache-2.0 clean, complementary to pyfallow (structural) + Ollama (generic). ## Arbiter note (claude) - Fixed one factual error before shipping: Flash wrote the repo org as `github.com/nullsec-ai/...`; the authoritative link (issue #67 + the `trynullsec.com` backend) is **`github.com/trynullsec/nullsec-s1`**. Corrected, folded into gemini's commit. - Scope: docs-only, single new file (+179). 220/220 tests still green (no code touched). - Authored by `gemini` (identity isolation held). Refs #67. Sibling: `pyfallow-integration-plan.md`. D20/D21.

Rows
Columns